ERN from EF bias
Hannah, 3/7/2017
Major changes: Changed models to cross-classfied models (grouping variables are subject and electrode, where electrode is not nested within subject.
Data from previously processed EEG files from CU and MU were used. Before I got them, the following was done:
- CNT file was merged with dat file to add response latency and accuracy information to CNT file
- CNT was re-referenced to average mastoids reference
- Blinks were corrected for
- A filter was applied (need to look up the settings for that)
- Files were response-locked, with an epoch of -400 to 500
I subsequently:
- Baseline corrected EEG files (using a baseline of -400 to -200)
- Performed an automatic artifact rejected procedure (trials with +- 75 uV were rejected, only using 9 electrodes of interest as the criteria)
Trials were included if:
1) The RT was between 200 and 500 ms
2) A response was made (i.e., no miss trials)
3) The trial wasn’t rejected in artifact rejection procedure
The following subjects were excluded:
- 1040 (doesn’t have full number of trials)
- 2023 (problems with EEG data)
- 2077 (problems with EEG data)
- 2089 (problems with EEG data)
- 2151 (problems with EEG data)
- 2157 (problems with EEG data)
- 2181 (problems with EEG data)
- 2187 (doesn’t have full number of trials)
Each subject did 384 experimental trials (prime-only trials were also included but not in the 384).
FZ, F3, F4, FCZ, FC3, FC4, C3, CZ, C4 (9 electrodes) were included.
Total sample is 134 subjects, 60 from CU and 74 from MU.
1. Examine correlation between subjects’ ERN mean amplitude and total number of errors
All conditions together:
## pearsons.r pvalue
## cor 0.08327981 0.3387364By condition separately:
## condition pearsons.r pvalue
## 1 Black-gun -0.028 0.745
## 2 Black-tool 0.244 0.004
## 3 White-gun -0.051 0.559
## 4 White-tool 0.104 0.230
## Saving 7 x 5 in image2. ERN/CRN grand averages
Negative is plotted upward.
To test the mean amplitude of the ERNs, a model was fitted with Race and Object as predictors, with ERN mean amplitude as the DV (correct trials are not included). The intercept, slopes of Race and Object, and their interaction were allowed to vary by subject. The intercept was allowed to vary by Electrode (not nested within Subject).
Race and Object were both effect coded.
Random effects:
## Groups Name Std.Dev. Corr
## Subject (Intercept) 2.87056
## Race.e 0.95851 -0.106
## Object.e 1.25871 -0.021 -0.003
## Race.e:Object.e 0.88495 -0.010 -0.198 -0.049
## Electrode (Intercept) 1.42298
## Residual 7.60126Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) -1.162 0.536 12.925 -2.167 0.049
## Race.e 0.638 0.088 128.250 7.250 0.000
## Object.e -0.518 0.113 129.674 -4.591 0.000
## Race.e:Object.e 0.218 0.082 124.040 2.653 0.0093. Looking at the ERN over the course of the experiment: Time-on-task
Slopes and estimates of lines are from the MLM, not fitted with OLS. Negative is plotted downward.
Simple slopes
Trial number is rescaled to range from 0 to 10.
## Estimate SE ci95_lower ci95_upper Race Object Color
## 1 0.00775 0.0211 -0.0345 0.050 Black gun light blue
## 2 0.11112 0.0168 0.0775 0.145 Black tool dark blue
## 3 0.08121 0.0182 0.0449 0.118 White gun light red
## 4 0.08662 0.0192 0.0482 0.125 White tool dark redModel output
The intercept, slopes of Race and Object and their interaction are allowed to vary by subject. The intercept is additionally allowed to vary by electrode (not nested within subject). Categorical variables are effect coded.
Trial is rescaled to range from 0 to 10. Each data point corresponds to ERN mean amplitude following an error, with the error’s original trial number on the x-axis. This spacing maintains time-on-task as the x variable. Fixed effects of Race, Object, and their interaction are estimated at the beginning of the task.
Random effects:
## Groups Name Std.Dev. Corr
## Subject (Intercept) 2.87040
## Race.e 0.95501 -0.101
## Object.e 1.25677 -0.022 -0.005
## Race.e:Object.e 0.88243 -0.004 -0.195 -0.057
## Electrode (Intercept) 1.42301
## Residual 7.59783Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) -1.5408 0.5385 13.1511 -2.8613 0.0132
## Race.e 0.5678 0.1013 227.0005 5.6060 0.0000
## Object.e -0.6583 0.1235 186.8736 -5.3300 0.0000
## Race.e:Object.e 0.3434 0.0962 236.0203 3.5689 0.0004## Estimate Std. Error df t value Pr(>|t|)
## Trial.s 0.072 0.009 86995.12 7.582 0.000
## Race.e:Trial.s 0.012 0.009 86925.74 1.295 0.195
## Object.e:Trial.s 0.027 0.009 87009.92 2.877 0.004
## Race.e:Object.e:Trial.s -0.024 0.009 86878.68 -2.592 0.0104. Looking at the ERN over the course of the experiment: Ordered errors
- Each error is ordered for each subject (i.e., error #1 for subject 10 and error #1 for subject 20 now have the same x-value, even if it happened on trial 12 for subject 10 and trial 50 for subject 20). Trial-type is ignored when ordering errors.
Slopes and estimates of lines are from the MLM, not fitted with OLS. Negative is plotted downward.
Simple slopes
Error order number is rescaled to range from 0 to 10.
## Estimate SE ci95_lower ci95_upper Race Object Color
## 1 0.0158 0.0332 -0.0506 0.0822 Black gun light blue
## 2 0.1383 0.0258 0.0867 0.1900 Black tool dark blue
## 3 0.0690 0.0283 0.0124 0.1257 White gun light red
## 4 0.0799 0.0295 0.0208 0.1389 White tool dark redModel output
The intercept, slopes of Race and Object and their interaction are allowed to vary by subject. Additionally, the intercept is allowed to vary by Electrode. Categorical variables are effect coded.
For each subject, error trials are numbered (in order) and rescaled to range from 0 to 10.
Random effects:
## Groups Name Std.Dev. Corr
## Subject (Intercept) 2.86538
## Race.e 0.95712 -0.101
## Object.e 1.25039 -0.029 -0.001
## Race.e:Object.e 0.88300 -0.005 -0.201 -0.044
## Electrode (Intercept) 1.42302
## Residual 7.59974Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) -1.3614 0.5374 13.0400 -2.5334 0.0249
## Race.e 0.6391 0.0965 186.7713 6.6210 0.0000
## Object.e -0.6050 0.1190 164.1297 -5.0855 0.0000
## Race.e:Object.e 0.2899 0.0911 190.3586 3.1824 0.0017## Estimate Std. Error df t value Pr(>|t|)
## TrialOrder.s 0.076 0.015 86455.18 5.127 0.000
## Race.e:TrialOrder.s -0.001 0.015 61586.13 -0.090 0.929
## Object.e:TrialOrder.s 0.033 0.015 75032.17 2.270 0.023
## Race.e:Object.e:TrialOrder.s -0.028 0.015 56110.16 -1.913 0.0565. Looking at the ERN over the course of the experiment: Ordered errors, centered by subject
- Each error is ordered for each subject, then centered for each subject
Slopes and estimates of lines are from the MLM, not fitted with OLS. Negative is plotted downward.
Simple slopes
Error order number is rescaled to range from -5 to 5.
## Estimate SE ci95_lower ci95_upper Race Object Color
## 1 0.0188 0.0336 -0.0483 0.0859 Black gun light blue
## 2 0.1315 0.0260 0.0794 0.1835 Black tool dark blue
## 3 0.0739 0.0285 0.0168 0.1309 White gun light red
## 4 0.0788 0.0299 0.0191 0.1386 White tool dark redModel output
The intercept, slopes of Race and Object and their interaction are allowed to vary by subject. The intercept is allowed to vary by electrode. Categorical variables are effect coded.
For each subject, error trials are numbered (in order), centered, and rescaled to range from -5 to 5.
Random effects:
## Groups Name Std.Dev. Corr
## Subject (Intercept) 2.87093
## Race.e 0.95693 -0.103
## Object.e 1.25753 -0.021 -0.003
## Race.e:Object.e 0.88474 -0.008 -0.197 -0.052
## Electrode (Intercept) 1.42302
## Residual 7.59973Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) -1.1630 0.5361 12.9218 -2.1692 0.0493
## Race.e 0.6358 0.0879 128.2830 7.2308 0.0000
## Object.e -0.5168 0.1128 129.6685 -4.5831 0.0000
## Race.e:Object.e 0.2155 0.0821 124.1305 2.6252 0.0097## Estimate Std. Error df t value Pr(>|t|)
## TrialOrder.c.s 0.076 0.015 86987.02 5.114 0.000
## Race.e:TrialOrder.c.s 0.001 0.015 86930.42 0.041 0.967
## Object.e:TrialOrder.c.s 0.029 0.015 86946.86 1.984 0.047
## Race.e:Object.e:TrialOrder.c.s -0.027 0.015 86916.94 -1.817 0.0696. Looking at the ERN over the course of the experiment: Ordered errors by trial type
- Each error is ordered for each subject, separately for each trial type
Slopes and estimates of lines are from the MLM, not fitted with OLS. Negative is plotted downward.
Simple slopes
Error order number is rescaled to range from 0 to 10.
## Estimate SE ci95_lower ci95_upper Race Object Color
## 1 0.0694 0.0607 -0.0520 0.191 Black gun light blue
## 2 0.1510 0.0310 0.0890 0.213 Black tool dark blue
## 3 0.0480 0.0396 -0.0312 0.127 White gun light red
## 4 0.1456 0.0481 0.0495 0.242 White tool dark redModel output
The intercept, slopes of Race and Object and their interaction are allowed to vary by subject. The intercept is allowed to vary by electrode. Categorical variables are effect coded.
For each subject, error trials are numbered (in order) separately for each trial type and rescaled to range from 0 to 10.
Random effects:
## Groups Name Std.Dev. Corr
## Subject (Intercept) 2.86338
## Race.e 0.95846 -0.102
## Object.e 1.24458 -0.034 0.006
## Race.e:Object.e 0.88777 -0.007 -0.195 -0.045
## Electrode (Intercept) 1.42301
## Residual 7.60001Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) -1.3382 0.5371 13.0144 -2.4917 0.0270
## Race.e 0.6609 0.0953 177.2619 6.9313 0.0000
## Object.e -0.6058 0.1175 158.7570 -5.1572 0.0000
## Race.e:Object.e 0.2366 0.0901 178.8324 2.6257 0.0094## Estimate Std. Error df t value Pr(>|t|)
## TrialOrder.type.s 0.103 0.023 82931.67 4.465 0.000
## Race.e:TrialOrder.type.s -0.007 0.023 69220.02 -0.290 0.772
## Object.e:TrialOrder.type.s 0.045 0.023 78733.78 1.937 0.053
## Race.e:Object.e:TrialOrder.type.s 0.004 0.023 67046.87 0.175 0.861